Correlation of PD-L1 Expression with Clinicopathological and Genomic Features in Chinese Non-Small-Cell Lung Cancer

Programmed cell death 1 ligand 1 (PD-L1) has been approved as predictive biomarker for non-small-cell lung cancer (NSCLC) patients treated with PD-(L)1 blockade therapy. The clinical/genomic features associated with PD-L1 are not well studied. Genomic profiling of tumor biopsies from 883 Chinese NSCLC patients was performed by targeted next-generation sequencing. Immunohistochemical analysis was conducted to evaluate PD-L1 expression levels using antibodies Dako 22C3 and 28-8, respectively. Our study showed distinct correlation between PD-L1 expression and clinical/genomic characteristics when using different PD-L1 antibodies and in different histological subtypes including adenocarcinoma (ADC) and squamous cell carcinoma (SCC), respectively. PD-L1 high expression (22C3) was associated with male and lymph node metastasis only in ADC patients. Furthermore, mutations of TP53 and KRAS, KIF5B-RET fusion, copy number gains of PD-L1 and PD-L2, and arm-level amplifications of chr.12p were significantly associated with PD-L1 positive status in ADC patients. For SCC patients, the gain of EGFR and MDM2 and loss of PTPRD were negatively associated with PD-L1 expression. We also compared our results with other studies and found conflicting results presumably because of the multiplicity of antibody clones and platforms, the difference of cutoffs for assigning PD-L1 expression levels, and the variation in study populations. Our study can help to understand the utility and validity of PD-L1 as biomarker of response to immune checkpoint inhibitors.


Introduction
Non-small-cell lung cancer (NSCLC) is the leading cause of cancer-related mortality worldwide, among which adenocarcinoma (ADC) and squamous cell carcinoma (SCC) are the most common histological subtypes [1]. Checkpoint blockade immunotherapy has had remarkable development in the last decade, and immune checkpoint inhibitor-(ICI-) targeted programmed cell death 1 (PD-1) and its ligand (PD-L1) have provided promising survival benefit for NSCLC patients [2].
To date, only PD-L1 expression, microsatellite instability (MSI), and tumor mutational burden (TMB) have been approved by the Food and Drug Administration (FDA) as predictive biomarkers for anti-PD-(L)1 therapies in patients with advanced NSCLC. However, it has been observed that MSI-high (MSI-H) or mismatch repair deficiency (dMMR) rarely appears in lung adenocarcinoma [3], and the predictive capacity of PD-L1 or TMB testing alone is insufficient [4][5][6]. Several effectors compromise the predictive accuracy of PD-L1 expression, including the heterogeneity of PD-L1 expression within tumor site, the multiplicity of antibody clones and platforms, and the difference of cutoffs for assigning PD-L1 expression levels [7]. Numerous studies have proved that the predictive value of PD-L1 expression is also affected by clinical, pathological, and molecular characteristics, such as age, sex, tumor site, and genomic alterations [8,9]. It has been reported that genomic alterations of KRAS, TP53, BRAF, and MET are associated with increased expression of PD-L1in NSCLC patients [10,11], while EGFR and STK11 mutations are correlated with negative/low PD-L1 expression [11,12], whereas genomic features related to PD-L1 are different among various study populations, antibody clones, and experimental platforms.
Since genetic testing is widely applied in genomic profiling to identify druggable alterations of NSCLC, we aimed to find the correlation between genomic alterations, TMB and PD-L1 expression, and whether genomic alterations can be used in discriminating PD-L1 level. Herein, we performed a retrospective study correlating the presence of clinical characteristics, genomic alterations, and PD-L1 expression in Chinese NSCLC patients including ADC and SCC, providing data of the association between PD-L1 expression and genomic features when using different PD-L1 antibodies and in different histological subtypes for identifying potential anti-PD-L1 treatment-related biomarkers in NSCLC.

Patients and Samples.
We retrospectively reviewed Chinese NSCLC patients who underwent next-generation sequencing (NGS) with a cancer-relevant gene panel of 139 or 425 genes (Supplementary Table S1) and PD-L1 testing on the same tissue sample at the Harbin Medical University Cancer Hospital, The Third Affiliated Hospital of Soochow University, The First Affiliated Hospital of Dalian Medicine University, and Cancer Hospital of the University of Chinese Academy of Sciences between November 2018 and July 2019. All NSCLC patients were pathologically confirmed. Exclusion criteria were poorquality tumor samples and incomplete clinicopathological features. The specimen slides less than 100 viable tumor cells or the percentage of tumor cells <10% were also excluded. The study was approved by the ethical committees of all participating hospitals. All patients have provided written informed consent. Overall, 883 specimens were collected and analyzed for clinicopathological characteristics including age, sex, histology type, pathological stage, and tumor site. All the samples were shipped to the central laboratory of a clinical testing center (Nanjing Geneseeq Technology Inc., China) for genetic and PD-L1 testing.

Next-Generation Sequencing and Bioinformatic Analysis.
Genomic DNA was extracted from tumor tissues, followed by sequencing library preparation according to published protocols [16]. Hybridization capture-based targeted NGS, which was used to selectively target 139 or 425 cancerspecific genes (clone 22C3: n 139 = 122 and n 425 = 628; clone 28-8: n 139 = 7 and n 425 = 126), was performed on the Illumina Hiseq platform (Illumina, San Diego, CA). The samples were analyzed as previously described to identify genomic alterations and were presented only when the percentage of tumor cells ≥10% [17]. Raw sequencing data were first demultiplexed by bck2fastq and then trimmed by trimmomatic as part of the quality control (QC) protocol [18]. The qualified reads were then mapped onto the human reference genome (GRCh37/UCSC hg19) using the Burrows-Wheeler Aligner (bwa-mem) [19]. PCR duplicates were removed using Picard suite (http://picard.sourceforge.net); base quality score recalibration and local realignment were performed using the Genome Analysis Toolkit (GATK, version 3.4) [20]. MuTect [21] and SCALPLE (http://scalpel .sourceforge.net) were then applied to identify mutations and structure variations (SVs). For oncogenic alterations calling, we used oncology knowledge base (OncoKB), a database for the oncogenic effects and treatment implications of cancer-related alterations, to identify candidate mutations and SVs in data from both the 139 and the 425 panels.
Copy number variations (CNVs) were only analyzed with data from patients subjected to the 425-panel sequencing, as the small number of cancer genes in the 139 panel provided challenges in accurately identifying CNVs. Genelevel and arm-level copy numbers were calculated with a reported pipeline [22,23]. We used a noise reduction model built from a pool of normal samples to process the read count from targeted regions of interest. A fold change threshold of 2 and 0.5 was used to identify gene CNV gain and loss, respectively. The arm-level CNV was identified if more than 50% of the corresponding chromosome segment length was either deleted or amplified.
2.4. Statistical Analysis. Statistical analyses for age were performed using the Wilcox test in group 1 (TPS = 1% cutoff for clone 22C3), group 2 (TPS = 50% cutoff for clone 22C3), 2 Journal of Oncology group 4 (TPS = 1% cutoff for clone , and group 5 (TPS = 10% cutoff for clone  and Jonckheere trend test in group 3 (TPS of <1%, 1-49%, and ≥50% for clone 22C3) and group 6 (TPS of <1%, 1-9%, and ≥10% for clone , respectively. We have defined the groups because PD-L1 IHC 22C3 is indicated as an aid in identifying NSCLC patients for different anti-PD-1/PD-L1 drugs, such as pembrolizumab (TPS ≥ 1%) or cemiplimab (TPS ≥ 50%) [24,25]. And PD-L1 IHC 28-8 is indicated as an aid in identifying NSCLC patients for treatment with nivolumab and atezolizumab (TPS ≥ 1%) [26]. Between-group differences of sex, metastatic sites and genomic alterations were examined using Fisher's exact test in groups 1, 2, 4 and 5 and the Cochran-Armitage test for trend test in group 3 and 6. As a minimal requirement, every mutation must be presented in at least six patients for each group. Correlations between TMB and PD-L1 (clone 22C3) expression were examined by using Spearman's rank correlation. The two-sided p values < 0.05 were considered statistically significant, and adjust p (adj. p) values with correcting for the false discovery rate ðFDRÞ < 0:25 were highlighted. Statistical analyses were performed with SPSS version 25.0 (IBM) software and R version 3.3.3 software.

Samples and Clinical
Characteristics. The clinicopathological characteristics of all 883 NSCLC patients were analyzed (Table 1) (Table 1). PD-L1 high expression (clone 22C3, TPS ≥ 50%) for ADC patients was more common in male (p = 0:006 in group 2) which was confirmed by binary logistic regression analysis with age and sex (p = 0:009, data not shown) and metastatic samples (p = 0:026 in group 2) (Figures 1(a) and 1(b)). In addition, there was no statistically significant relationship between PD-L1 expression and tumor stage (data not shown). In the SCC patients, lymph node metastasis was enriched in PD-L1-positive group (clone 22C3, TPS ≥ 1%) compared to samples with other metastatic sites (p = 0:031 in group 1) (Figure 1(c)). In clone 28-8-detected PD-L1 level, no significant association between PD-L1 expression and the clinicopathological characteristics was identified for ADC patients. There were only 14 SCC patients who had PD-L1 testing by clone 28-8, so we could not analyze their features with PD-L1 expression.

Association between Oncogenic Genomic Alterations and PD-L1 Expression.
Further study of the association between genomic alterations and PD-L1 expression (clone 22C3) in ADC showed that the mutations of TP53, EGFR, KRAS, RET, and POLE were the most relevant to PD-L1 expression ( Figure 2). The Cochran-Armitage trend test showed that PD-L1 expression (clone 22C3) was positively associated 3 Journal of Oncology with TP53 and KRAS and negatively associated with EGFR, RET, and POLE.
As oncogenic mutations may have more direct functions in tumorigenesis, we then used a gene list from OncoKB to search for the oncogenic alterations in our data and observed that oncogenic mutations of EGFR, KRAS, and TP53 were significantly related with PD-L1 expression (clone 22C3) in ADC (Figure 2 middle). We then focused on the difference between oncogenic mutations of EGFR L858R and exon 19 in-frame deletion (19Del) in association with PD-L1 expression, as they are the two most well-studied oncogenic mutations showing distinct functions in mediating carcinogenesis [27,28]. Interestingly, the frequency of cases with TPS < 1% was higher in EGFR 19Del than EGFR L858R (63% vs. 51%). Meanwhile, PD-L1 positive (TPS ≥ 1%) was enriched in patients with EGFR L858R but significantly inversely associated with EGFR 19Del (p = 0:021 and adj:p = 0:047 in group 2, Supplementary Table S2). Furthermore, our results suggested that the association between PD-L1 expression and other EGFR oncogenic mutations (EGFR other ) such as exon 20 insertion and T790M was similar to EGFR 19Del . Additionally, all the oncogenic mutations of KRAS (p < 0:001 and adj:p = 0:004 in group 3) and especially KRAS G12 which was the most common KRAS oncogenic mutation (p = 0:002 and adj:p = 0:007 in group 3) had higher mutation rates when PD-L1 expression (clone 22C3) was higher (Supplementary Table S2).
Interestingly, POLE mutation was enriched in the PD-L1-positive/ PD-L1-high (clone 22C3) group in SCC, but not in the PD-L1-negative group in ADC ( Figure 2). As an enzyme involved in DNA repair, POLE mutations have been associated with disruption of the exonuclease activity required for proofreading function, which results in a high TMB level and is vulnerable to ICIs [29,30]. In our data, the different correlation of POLE with PD-L1 expression in SCC and ADC indicated distinct responses of SCC and ADC patients with POLE mutations upon ICI treatment. Further studies should be conducted to validate this hypothesis. Additionally, oncogenic TP53 mutations exhibited lower frequency in the SCC population with positive PD-L1 expression (clone 22C3, p = 0:049, Supplementary  Table S2) but not in the ADC population. In SCC, PD-L1 expression (clone 22C3) was positively associated with ALK and NFE2L2, and negatively associated with APC ( Figure 2).
As TMB is used as an independent biomarker for immunotherapy, we further analyzed the relation between TMB   Table S3). PD-L2 is also a ligand of PD-1, which mediates T cell activity inhibition [31]. It has been reported that PD-L2 expression was correlated with PD-L1 in esophageal squamous cell carcinoma [32], indicating a functional relation of PD-L1 and PD-L2 in tumorigenesis. Therefore, we focused on PD-L2 and found its copy number gain to be significantly in agreement with PD-L1 expression, implying an interaction  Table S4). For the 119 SCC patients, gain of EGFR and MDM2 (p = 0:040 and adj:p = 0:234 and p = 0:044 and adj:p = 0:234, respectively) and loss of PTPRD (p = 0:015 and adj:p = 0:234) were negatively associated with PD-L1 expression (Supplementary Table S3). It has been reported that patients with MDM2 amplification have higher prevalence of hyperprogression when treated with anti-PD-(L)1 immunotherapy [33]. Additionally, our data showed a significantly inverse correlation between MDM2 copy number gain and PD-L1 expression (clone 22C3) in both ADC and SCC patients, which indicated a functional regulation of PD-L1 by MDM2 in NSCLC. PD-L1 expression (clone 22C3) was negatively associated with copy number amplification in chr.14q (p = 0:032) and chr.20q (p = 0:026) and deletion in chr.19p (p = 0:025) but positively associated with chr.9p amplification (p = 0:008 and adj:p = 0:211) and chr.13q deletion (p = 0:019) (Supplementary Table S4).

Discussion
We retrospectively studied PD-L1 expression and genomic alterations to identify the correlation between them in Chinese NSCLC patients. Since each ICIs has its own antibody and clinical cutoffs, which was specially developed and associated with different clinical trials, we divided the patients accordingly into two cohorts (clone 22C3 or clone . In our study, higher PD-L1 expression (clone 22C3) in male ADC patients was identified, which was consistent with the previous studies [8,34]. Our data also showed that PD-L1 expression (clone 22C3) was significantly higher in tissue derived from a metastatic site compared to the primary tumor in ADC patients. This finding suggests that PD-L1 expression may be higher in advanced disease than in earlier stages [35]. Additionally, the proportion of cases with PD-L1 high expression (clone 22C3) among metastatic sites is higher in lymph node than other sites in SCC patients but not in ADC patients. PD-L1 expression may vary among different tumor sites, indicating that repeat biopsy and PD-L1 staining can be conducted to improve the predictive capacity of PD-L1 [36].
We also observed genomic alterations correlating with PD-L1 expression (clone 22C3). In the ADC population, alterations of TP53 and KRAS were positively associated with PD-L1 expression. In contrast, EGFR and RET alterations were associated with PD-L1 negatively. The association between PD-L1 expression (clone 22C3) and oncogenic alterations from OncoKB provides information on the prognostic and predictive significance of somatic alterations, which can be used to optimize treatment decisions [37]. For further investigation of oncogenic alterations, we also find the difference between EGFR 19del and EGFR L858R , as PD-L1 positive was enriched in patients with EGFR L858R but inversely associated with EGFR 19Del . Conversely, all the KRAS oncogenic mutations and especially KRAS G12 are correlated with high PD-L1 expression (clone 22C3). Interestingly, TP53 oncogenic mutations have no association with PD-L1 expression in ADC patients but are associated with negative PD-L1 expression (clone 22C3) in SCC patients. Previous study has shown that TP53 mutations affected immune checkpoints expression, T cell infiltration, and tumor immunogenicity in lung ADC [38]. Besides, a research has shown that TP53 oncogenic mutations were enriched in PD-L1-high group of American patients with nonsquamous NSCLC which was different from our study [12]. Moreover, other studies showed a correlation between TP53 mutations and PD-L1 expression as well as response to ICIs in ADC [39] or SCC [10]. Since TP53 mutations increased numbers of somatic mutations and expression of neoantigens, high TMB is more likely to benefit from immunotherapy [10,40]. The distinct correlation between PD-L1 and TP53 mutations may be due to the difference between populations, experimental platforms, and methods of PD-L1 testing. It has been reported that amplifications in MDM2 and MDM4 are associated with hyperprogression with anti-PD-(L)1 therapy [38]. In our study, MDM2 amplification is correlated with low PD-L1 expression (clone 22C3). In our data, the different correlation of POLE with PD-L1 expression in SCC and ADC may indicate distinct responses of SCC and ADC patients with POLE mutations upon ICI treatment.
There was no significant association between PD-L1 expression and the clinicopathological characteristics in ADC patients detected with PD-L1 clone 28-8. And the genetic alterations related to PD-L1 were different in clone 22C3 and clone 28-8. For PD-L1 antibody clone 28-8detected ADC patients, NKX2-1 gain and deletions of chr.9p were negatively correlated with PD-L1 expression. Additionally, deletions of chr.19p and 19q were significantly enriched in PD-L1-low group. Interestingly, no mutation was found to be related with 28-8-detected PD-L1 expression. These differences may be caused by the limited sample size of patients with 28-8-detected PD-L1 level (n = 93) and the different epitopes and cutoffs of 28-8 and 22C3 as the previous studies [41,42].
Genomic features related to PD-L1 are different among populations and experimental platforms. We have reviewed previous studies regarding the association between molecular features and PD-L1 expression using different antibodies ( Table 2). Using the SP142 antibody with the cutoff of ≥1% and ≥50% for TPS, and at ≥1% and ≥10% for immune proportion score (IPS), PD-L1 was correlated negatively with EGFR mutations and positively with KRAS, BRAF, and MET mutations and ROS1 translocations, in Chinese NSCLC patients [11]. On the other hand, using a rabbit polyclonal anti-PD-L1 antibody and a median histological score value of 30 as the cutoff point, the presence of EGFR mutations was found to be associated with increased PD-L1 expression in Japanese NSCLC patients [43]. It also reported that PD-L1 expression was associated TP53, KRAS, and STK11 mutations when using the primary antibody 5H1 with the cutoff at ≥1% in Germany NSCLC patients [10]. Using different antibodies, including E1L3N, 22C3, 28-8, SP142, and SP263, PD-L1 subgroups were defined as negative (PD − L1 < 1%), intermediate (PD-L1 1-49%), and high (PD − L1 ≥ 50%), and mutations in KRAS, TP53, and MET were demonstrated to be associated with PD-L1 high expression and STK11 mutations associated with PD-L1 negativity in American patients with lung adenocarcinomas [44]. In addition, in a Chinese NSCLC cohort, it was also confirmed that the TP53/KRAS subgroup manifested exclusive increased expression of PD-L1 and a highest proportion of PD-L1 + /CD8A + [39]. Furthermore, PD-L1 positivity was correlated with copy gain of CD274 (PD-L1) and PDCD1LG2 (PD-L2) in American nonsquamous NSCLC patients in a study using clone E1L3N with the cutoff at ≥1% [12]. Moreover, 11q13 amplification was found to be associated with high PD-L1 expression in Chinese NSCLC patients in a study using the 22C3 and 28-8 antibody at TPS cutoff values of ≥1% [45]. Recent studies have reported that PD-L1 expression is associated with STK11 mutations, KEAP1 mutations, APC mutations, and JAK2 gain [12,46], which was not observed in our study. These conflicting results are presumably because of the multiplicity of 6 Journal of Oncology antibody clones and platforms, the difference of cutoffs for assigning PD-L1 expression levels, and the variation in study populations. TMB and PD-L1 expression are the two independent predictive biomarkers for anti-PD-1/PD-L1 therapy [34]. In line with the previous findings, we find no association between TMB and PD-L1 expression [44,47]. Our model, which used a combination of seven markers, including EGFR oncogenic mutation, KRAS oncogenic mutation, PD-L2 gain, PD-L1 gain, MDM2 gain, chr.1q amplification, and chr.20q amplification, showed good performance to associate PD-L1 expression (TPS = 50% cutoff for clone 22C3) in ADC patients. Based on the model results, patients with PD-L1 high had a significantly longer PFS compared to low PD-L1 expression. These results indicated that the model can help to understand PD-L1 level for patients who have no tissue available for PD-L1 IHC.
There are still several limitations of our study which should be considered and further studied in the future. Firstly, not all enrolled patients provided completely clinical characteristics, especially the smoking status, the stage, and the treatment history. Previous study observed an association of PD-L1 high expression with smoking status [48,49] which may be related to the alteration of tumor microenvironment by smoking [9], whereas no smoking data was available in our study which may be regarded as an issue of the real-world data [50,51]. Additionally, we did not observe a statistically significant relationship between PD-L1 expression and tumor stage in the ADC which may due to the small sample size of patients with stage I-II lung cancer in our study. Further research is warranted to study the association between PD-L1 expression and patients' clinico-pathological characteristics. Secondly, our model only included patients with clone 22C3-detected PD-L1 expression with the 50% cutoff, and researches on different PD-L1 antibodies and their cutoffs were still needed. Furthermore, a large population and more genomic features should be studied in order to further understand which patients may respond to ICIs.

Conclusions
This study revealed the correlation between PD-L1 expression, clinical features, genomic alterations, and TMB in Chinese NSCLC patients and highlighted the discordance of the association between PD-L1 expression and genomic features when using different PD-L1 antibodies and different histological subtypes including ADC and SCC. PD-L1 subgroups were defined different groups according to the cutoffs of approved treatment for NSCLC patients. Moreover, our study is comprehensive and expensive since our analysis uses a large database of more than 880 Chinese NSCLC cases, and we compared our results with other studies and found conflicting results. Our results help to understand the relationship between genomic alterations and PD-L1 expression and may provide a novel idea for application of molecular features.

ADC:
Adenocarcinoma SCC: Squamous cell carcinoma TMB: Tumor mutational burden Tumor proportion scores NGS: Next-generation sequencing QC: Quality control SVs: Structure variations CNVs: Copy number variations OncoKB: Oncology knowledge base FDR: False discovery rate ROC: Receiver operating characteristic DCB: Durable clinical benefit NDB: No durable benefit HR: Hazard ratio.

Data Availability
All data relevant to the study are included in the article or uploaded as supplementary information.

Conflicts of Interest
XH, YJ, SL, XX, YS, WT, QO, HB and XW are the shareholders or employees of Nanjing Geneseeq Technology Inc., China. Other authors have no conflicts of interest to declare.

Authors' Contributions
YL, YZ, and YW conceptualized the study. CL, HB, and XX were responsible for the methodology. XW, YS, and MX carried out the investigation. YJ and XH wrote the original draft. SL, WT, and QO wrote, reviewed, and edited the manuscript. YW was responsible for the funding acquisition. YZ and YW supervised the study. All authors read and approved the final manuscript. Yue Li and Chong Li contributed equally to this work.

Acknowledgments
The work was supported by the Zhejiang Natural Science Foundation (20160171).

Supplementary Materials
Supplementary Figure S1: representative images of immunohistochemical staining for PD-1 with 22C3 and 28-8 from ADC or SCC patients.